联合PolSAR和InSAR的湿地变化监测研究

The time series SAR data can truly reflect the changes that have occurred in the same geographical area for a long time. As a complex ecosystem, wetlands range from permanently flooded areas to lands that are floods or saturated for extended periods usually with some frequency. In this proposal, the main research focuses on how to monitor periodic and large-scale changes in wetlands. During recent years, there have many methods of monitoring wetlands using the SAR images. However, these methods still have many problems, such as inadequate utilization of polarization information, inadequate accuracy of extracting ground scattering mechanism, false alarm and missed detection in change detection map, and inability of extracting change information from vertical structure. In this proposal, machine learning and SAR technologies are used to solve these problems above. The method based on deep learning is used to obtain optimization features of wetlands. Moreover, the joint classification change detection method based on ensemble learning obtains the changes of types of land cover accurately. Finally, the method combined PolSAR and InSAR technologies extract large-scale water level changes in wetlands. This proposal has been developed by detecting the changes in planar structure and vertical structure of wetland hydrology. It provides scientific and technological support for accurate interpretation of wetland features, ecological environment protection, regional economic development planning and sustainable development.

时间序列SAR影像能够真实反映地物在一个长时间范围内的动态变化情况，湿地作为水陆互相作用形成的复杂生态系统，如何对其进行周期性、大范围变化监测是本课题的研究重点。针对现有利用SAR影像进行湿地监测方法存在极化信息利用不充分、地物散射机制提取精度不足、变化检测易产生虚警和漏检、检测结果无法提取地物垂直结构变化信息等问题，本课题拟利用深度学习和集成学习方法，联合PolSAR和InSAR技术，开展基于深度学习的PolSAR湿地散射特征集选取方法、基于集成学习联合分类的多时相PolSAR湿地变化检测方法和联合PolSAR和InSAR技术的大范围湿地水位提取方法等关键技术的研究，从而实现湿地复杂地物类型平面结构变化检测及湿地水文垂直结构动态监测，形成时序SAR影像对湿地时空动态监测技术，为湿地地物精确解译、生态环境保护、区域经济发展规划和可持续发展等应用提供了科学技术支撑。

湿地作为水陆互相作用形成的复杂生态系统，有效开展湿地动态监测是湿地保护的重点方向，是掌握湿地演化的重要支撑。针对利用SAR影像大范围湿地动态监测过程中，存在极化信息利用不充分、湿地地物散射特性提取精度不足、变化监测易产生虚警和漏检、监测结果无法估计地物垂直结构变化等问题，本项目以“深度学习特征优化-动态变化监测-大范围水文结构提取”为研究思路，分别开展基于深度学习的PolSAR湿地散射特征集选取方法、基于集成学习联合分类的多时相PolSAR湿地变化检测方法、联合PolSAR和InSAR技术的大范围湿地水位提取方法等关键技术的研究，完成了国内外包括：我国额尔古纳湿地、武汉湿地、黄河三角洲湿地、潘安湖湿地、鄱阳湖湿地和美国路易斯安纳湿地研究区数据汇总及实验，得到以下重要结论：1）本研究方法构建的优化特征集包含更低的信息冗余和更稳定的信息表达能力，且不依赖人工经验，有助于提升后续变化检测模型的收敛速度和目标提取精度。2）本研究方法解决了非监督变化检测不能检测地物类别变化，检测结果存在虚警和漏检；监督变化检测依赖分类精度，忽略时间信息的问题，扩展并提高了湿地地物变化检测的能力和精度。3）联合PolSAR和InSAR两种主流技术的优势，对湿地水域边界进行精确提取，进而实现湿地复杂地物类型平面结构变化检测以及湿地水文垂直结构动态监测。4）本项目关键技术已被证明能够应用于洪涝灾害和矿区监测等场景，并取得了不错的监测效果。本研究促进PolSAR与InSAR技术的融合，提供更为丰富的湿地动态监测产品，精确掌握湿地水位变化与周围环境的联动关系，进而为湿地保护提供强有力的数据和技术支撑，并将进一步扩展SAR技术理论，为PolSAR-InSAR在其他监测领域应用提供参考价值。

内容获取失败，请点击重试